Historical Sequential Transformers for AI-augmented Diagnostics (HIST-AID). This model implements a multi-modal architecture designed to replicate the clinical workflow of a radiologist by integrating current visual evidence with longitudinal medical history.
The architecture utilizes a Vision Transformer (ViT) for image feature extraction and a BERT-Base encoder for processing sequential radiology reports. These representations are fused via a Transformer-based fusion layer using cross-modal self-attention to generate context-aware diagnostic predictions.
.. autoclass:: pyhealth.models.hist_aid.HistAID
:members:
:undoc-members:
:show-inheritance: